Microsoft just published its 2026 Secure Access report, and the numbers are blunt: 97% of organizations experienced an identity or network access incident in the past year. 70% of those incidents involved AI-related activity — things like AI-assisted phishing and agent privilege escalation.

This isn’t a vendor pitch. It’s empirical data that confirms what the security community has been warning about for months: AI agent adoption is outpacing governance, and the gaps are already being exploited.

The Scale of the Problem

The report’s key findings paint a picture of enterprises that have lost visibility into their own access environments:

  • 97% experienced identity or network access incidents in 12 months
  • 70% of incidents tied to AI activity (phishing, agent escalation, shadow AI)
  • 53% of incidents were malicious; 47% were accidental
  • 6 in 10 leaders expect more incidents specifically from AI agents and GenAI tool use
  • Average organization uses 5 identity solutions and 4 network access solutions from different vendors
  • 40% say they have too many vendors; 32% say their tools are duplicative

That near-even split between malicious and accidental incidents is telling. Half the problem isn’t adversarial attacks — it’s complexity, unclear ownership, and misaligned controls. When employees adopt AI tools and teams deploy agents faster than policies can be updated, unintentional misuse becomes inevitable.

Why Traditional IAM Breaks

Microsoft frames the core challenge as a fundamental mismatch between how identity systems were designed and how AI agents actually work:

Traditional IAM assumes:

  • Humans log in once
  • Access decisions are made at authentication time
  • Session-based trust persists until logout
  • Identity providers manage a bounded set of users

AI agents do this:

  • Operate continuously, often 24/7
  • Interact with multiple systems simultaneously
  • Require broad access to function as intended
  • Create new identity types that don’t fit existing categories
  • Generate access patterns at machine speed

Every AI tool, integration, or agent introduces new identities, permissions, and pathways. These identities don’t behave like traditional users. When that disconnect meets fragmented tooling — five identity solutions and four network access solutions on average — gaps emerge that are invisible to any single tool.

The “Access Fabric” Response

Microsoft’s proposed architectural answer is what it calls an access fabric — not a product, but an approach that treats access as a continuous, end-to-end system:

  • Common identity foundation for employees, workloads, and AI agents
  • Real-time enforcement of access decisions across the network
  • Continuous signal sharing across identity, network, and security tools
  • Policy propagation without manual stitching between systems

The key insight: AI systems and automated attacks operate at machine speed. Static access decisions and delayed enforcement create gaps that are difficult to detect and harder to close. The response has to be equally fast.

64% of organizations say they’re already consolidating identity and network access tools. 94% prefer a comprehensive, integrated IAM platform over best-of-breed point solutions. The vendor sprawl era is ending — not because integrated platforms are better in theory, but because the agent era makes fragmentation genuinely dangerous.

The RSAC 2026 Context

This report drops three days before RSAC 2026, where agent identity security has emerged as the dominant theme. In the past two weeks alone:

  • Oasis Security raised $120M for Agentic Access Management
  • 1Password launched Unified Access with Anthropic/OpenAI partnerships
  • Okta announced agent identity controls (April 30 launch)
  • SailPoint × AWS signed a multi-year deal for agent identity governance
  • F5 introduced NGINX Agentic Observability for MCP traffic inspection
  • Ping Identity released Identity for AI — single control plane for agent lifecycle
  • ConductorOne, Entro, Apono launched agent-specific governance products

Microsoft’s data validates what all these companies are building toward: the identity layer is the critical control surface for the agent era, and most enterprises aren’t ready.

What OpenClaw Users Should Know

If you’re running OpenClaw agents, you’re operating in the same environment Microsoft’s report describes. Your agents create machine identities, consume credentials, and make access decisions — often across multiple systems per task.

The 47% accidental incident rate is the number to watch. Most agent-related security issues won’t come from sophisticated attacks. They’ll come from:

  • Over-permissioned agents that inherit your full access instead of scoped credentials
  • Unmonitored tool calls where you can’t see what your agent accessed
  • Persistent credentials that never expire or rotate
  • Shadow agents — tools your team adopted without security review

The fix isn’t complex: scope credentials, audit access, rotate tokens, and know what your agents are doing. The hard part is doing it consistently as the number of agents grows.

Microsoft’s 2026 Secure Access report was published March 20, 2026. RSAC 2026 runs March 23–27 in San Francisco.